Apollo Economist Warns AI Agents Could Pull Cheap Deposits From Banks

Apollo Chief Economist Torsten Slok warned in a Sunday research note that AI assistants managing household money could set off a slow-motion bank run. Slok argued that agents could automatically move idle cash out of low-interest checking accounts and into higher-yielding alternatives, draining a funding source banks rely on to make loans.

Slok Describes a Slow-Motion Deposit Run

Slok titled the note “Is an Agentic bank run coming?” He wrote that Meta’s personal agent Muse and similar AI assistants could soon sweep household cash into high-interest accounts automatically. The national average interest rate on checking accounts is about 0.1%.

Slok is a widely read economist on Wall Street and a partner at Apollo Global, which manages about $1 trillion in assets. He said the scenario would put banks at risk because they depend on cheap deposits to create credit.

“Banks could lose a large share of the cheap deposits.”

Slok added that the loss of that funding would be a problem for the entire financial system if every household used AI agents to optimize the return on their cash balances. The note describes a hypothetical outcome of widespread adoption, not a current trend in deposit data.

The Rate Gap Behind the Argument

Slok pointed to firms including Revolut, SoFi, Varo, LendingClub and Wealthfront that pay between 3.3% and 5% annual interest on deposits. On a $10,000 balance, that range works out to $330 to $500 a year. The same balance earns about $10 a year in a checking account paying 0.1%.

That gap already exists today for consumers who choose to move their money, and many households have not acted on it. Slok’s argument is that agents remove the friction that keeps most households from acting on it. An agent could monitor balances continuously and shift cash without the account holder needing to compare rates or fill out new applications.

What Agentic Finance Refers To

Agentic finance describes AI that takes actions rather than only answering questions. These agents can track balances in real time, compare returns across institutions, move idle cash into higher-yield accounts and move it back in time to cover bills.

Estimates of the market’s size vary widely. Mordor Intelligence puts agentic AI in financial services at $7.78 billion in 2026 and projects $43.52 billion by 2031. MarketsandMarkets sizes the narrower AI agents segment at about $845 million in 2025. The gap reflects differences in how each firm defines the category.

Crypto Payment Rails Already Serve Agents

Crypto infrastructure is already being built for this type of software. Coinbase’s x402 protocol, described as the most widely used agentic payment standard, lets an AI agent pay for online services in stablecoins within seconds, with no account, card or human approval.

The protocol has reportedly processed roughly 188 million to more than 205 million cumulative transactions, with about 69,000 active agents, according to a third-party tracking site. Cloudflare, Google, Visa, Mastercard, AWS, Circle and Stripe have joined the x402 Foundation, which the Linux Foundation now governs.

Nate Geraci, co-founder of the ETF Institute, has previously said that AI and crypto are both coming for the traditional banking model. His comment predates Slok’s note.

What the Note Does Not Show

Slok’s warning rests on adoption that has not occurred at scale. Agent-driven transfers of household cash are not yet visible in bank deposit figures, and the x402 activity described above involves payments for online services rather than consumer savings accounts. The note also does not estimate how much in deposits could move or over what period.